2022 SOST Opportunities and Actions Roundtable Submissions
Summary: The world produces 179 million tonnes of seafood per year with a total estimated value of over $400B USD. Global production of seafood has quadrupled in the past 50 years leading to a third of global fish stocks being over-exploited, resulting in substantial losses in biodiversity and fisheries collapses. Managing fisheries and natural wildlife for long-term sustainability requires knowledge of fishing mortality, population recruitment, and population abundance to inform stock assessments and ensure that catch limits do not exceed overfishing thresholds.
Fishery assessments rely heavily on direct human observation – invasive catch and release measurements, divers conducting censuses of fish populations, and manual annotations of underwater videos – and require significant investments and science/management capacity that often do not exist, especially in the thousands of small-scale, data-limited fisheries operating around the world.
It is crucial to modernize fisheries assessment techniques to be cost effective, accurate, non-extractive and non-destructive, and capture a large portion of the population.
FishSense brings together recent innovations in GPU compute platforms, 3D depth cameras, and ML software to modernize fish stock assessments. FishSense technologies include an underwater RGBD camera and associated software analysis framework. FishSense integrates an nVidia embedded GPU and an Intel RealSense 3D camera into an underwater housing for use in scientific diving, aquaculture farms, fishing boat decks, underwater vehicles, and other fish assessment scenarios. FishSense utilizes scientific advancements in underwater imaging to collect underwater RGBD video and leverages ML on GPUs to analyze that data to provide meaningful assessment of fish health and populations.
Sector: Academia
Organization: UC San Diego
POC: Ryan Kastner, kastner@ucsd.edu